Robotic Implantation of Intracerebral Electrodes forDeep Brain Stimulation

C. Faria1∗, E. Bicho1∗, M. Rito3 , L. Louro1, S. Monteiro1 and W. Erlhagen2

1Department of Industrial Electronics and Center Algoritmi, University of Minho, Portugal2Department of Mathematics and Applications and Center of Mathematics, University of Minho, Portugal

3Service of Neurosurgery, Coimbra University Hospital, Portugal∗cfaria@dei.uminho.pt; estela.bicho@dei.uminho.pt

Abstract—This dissertation objective is to contribute for thedevelopment of a robotic system towards neurosurgery assistancein Deep Brain Stimulation (DBS) stereotactic procedures. BeingDBS neurosurgery typically a long, physically and cognitivelydemanding procedure; the introduction of a robotic assistant tohold, manipulate and position instrumentation would improvethe medical team working conditions and lead to better surgeryoutcomes. Upon understanding how could the robot be used andwhat robotic systems were adequate to the task, we implementeda simulation environment to emulate several industrial robotmanipulators and the operating room. It was also developedeach robot geometric and differential kinematic equations, andcontrol algorithms specifically oriented for DBS neurosurgeryassistance. Taking into account the operating room arrangement,the robot characteristics and task requirements, we selected themost apt industrial robotic manipulator and further elaboratedon its placement and orientation to achieve utmost performance.

Index Terms—Robotic neurosurgery; Stereotactic electrodeplacement; Deep Brain Stimulation (DBS)

I. INTRODUCTION

Deep Brain Stimulation is a technique used in functionalneurosurgery to stimulate specific basal ganglia regions viaimplanted electrodes. These electrodes are connected to a neu-ropacemaker that generates precise and controlled electricalsignals [1]. DBS treatment (with different stimulation param-eters) relieves symptoms of neurological disorders, rangingfrom: Parkinson’s disease, epilepsy, dystonia, essential tremorand even psychiatric related pathologies. Recent epidemiologystudies show the tremendous and growing impact of each ofthese pathologies in today’s society [7] [9] [13].

The DBS treatment has earned a strong reputation amongthe neurologist community, due to the resulting symptomaticrelief achieved, through a reversible procedure with few knownside-effects. On the other side, the standard DBS surgerycourse of action meets several repetitive, iterative and time-consuming – yet precision demanding steps [11] [14]. Thuscomes the main motivation for our dissertation work, which

Master ThesisPortuguese chapter of IEEE EMBS3rd Portuguese Meeting in Bioengineering, February 2013University of Minho

involves the development of a robotic system to perform theserepetitive and menial tasks in collaboration with the medicalteam. The projected neurosurgery robot, or as we like to callit, an ’intelligent surgical tool’ should therefore: place, holdor precisely manipulate other surgical instrumentation and alsointegrate safety check routines.

Robotic technology continuous progress, have brought itsprecision and repeatability standards to a limit that far ex-ceeds human’s capabilities [3]. However, artificial intelligencecan not compete with neurosurgeon’s dexterity, judgementexperience among other advantages [2]. Our aim is to bringforth the best of both worlds, in other to improve the medicalteam working experience and potentially improve the surgeryoutcome, through precise and consistent methodologies. Fur-thermore, we found that the price of an entire robotic systemis roughly a limiting factor, since it is about half the cost ofa simple mechanical not-actuated stereotactic device used inDBS surgery simply to position electrodes.

Facing the extremely challenging task of developing anentire robotic system oriented to surgery in the context ofa master thesis, we set as objective: to contribute for thedevelopment of a robotic system towards neurosurgeon’s as-sistance in DBS procedures. We took the first steps towardsunderstanding: what should the robot do in a DBS neuro-surgery; what is already done and what can be improved inneurosurgery robots; and developing an initial virtual roboticsolution not only to test several robot systems but also toaddress anticipated real implementation issues.

In section II it will be presented the course of action ofa standard DBS neurosurgery and will be emphasized howshould the robot be of use; in section III we will summar-ily present the state of the art in robotic neurosurgery forstereotactic procedures. In section IV, we will present thechoice process of the industrial robotic systems along withthe developed kinematic equations for each; and in sectionV the virtual tool created to test the robotic systems and theimplemented control algorithms. To conclude in section VI,it will be presented the first results and in section VII wewill describe the conclusions of this dissertation, and pointout future work.

II. DEEP BRAIN STIMULATION

The first task of the dissertation was to understand therobot utility in a DBS neurosurgery. To do so, we started byattending to a standard DBS neurosurgery that took place inCoimbra University Hospitals (Portugal), and was conductedin a Parkinson’s disease affected patient.

A. Neurosurgery ProcedureAfter being selected for DBS neurosurgery, the patient

undergoes imaging scans to pinpoint the brain structures tobe stimulated, as well as the skull entry coordinates throughwhich, each electrode will pass to reach the stimulation coordi-nates. Both coordinates are extrapolated based in a stereotacticreference frame1 (fig. 1a) that is previously fixed to thepatient’s skull. Along the patient’s preparation, the coordinatesof the stimulation target are simulated via a phantom device(fig. 1b). The mechanical screws of the stereotactic framemounted in the phantom, are set to position a driver systemthat will guide the electrodes past the skull entry point towardsthe stimulation target.

(a) Stereotactic Reference System. (b) Surgical coordinates verification.

Fig. 1. Intraoperative surgery preparation steps.

Once verified the phantom and frame coordinates for onestimulation target, the stereotactic frame is dismounted fromthe phantom and attached to the reference system on thepatient’s head (fig. 1a). The stereotactic frame is once againset to mark the skull entry point, and then moved aside to clearthe neurosurgeons’ workspace. With the skull entry point, theneurosurgeon team proceeds to incise the scalp and drill a burrhole in the patient’s skull.

The stereotactic frame is positioned one final time to assistthe neurosurgeons descend several sets of electrodes towardsthe stimulation target (to register brain activity or to applystimulation signals). Once the electrodes are placed, the stim-ulation variables like: electrode depth and signal intensity areadjusted, according to the variation of the patient’s symptoms2.

If the DBS surgery is bilateral, all the procedure must bethoroughly repeated for another stimulation target.

B. Robotic System UtilityBased on the information gathered about DBS surgery, we

thought of several potential improvements to the standardsurgery with the introduction of a robotic system:

1In this case, it was a ring-shaped structure.2The patient sedation is lowered and the patient remains awake during the

symptoms evaluation by neurophysiologists.

1) Interface the imaging software – where the surgerycoordinates are generated – with the robotic controller;

2) Avoid mount/dismount and the process of setting thestereotactic frame coordinates, by making the procedure’frameless’;

3) Assist the neurosurgeon in skull drilling, by constrainingthe trepan trajectory;

4) Swiftly position and manipulate instrumentation withguaranteed precision and consistency;

5) Assist highly experienced senior neurosurgeons thatmight lack the required dexterity; or guide young andmuch less experienced neurosurgeons.

III. STATE OF THE ART

Having outlined the robotic system goal, it was essential toperform a state of the art search on current neurosurgical robotsolutions oriented to stereotactic brain surgeries. We searchedon MeRoDa (medical robotics database) and in several reviewpapers, for robotic systems that fall in the above cited categoryand found some oriented and potentially adaptable systems [5][12] [6].

Among the robotic systems oriented to minimally invasivestereotactic neurosurgery we selected: i) Neurobot, developedin Imperial College of Science, Technology and Medicinein London; ii) NeuroMate, by Renishaw; iii) Pathfinder, byProsurgics Ltd.; iv) Robocast, from Neuroengineering andmedical robotics laboratory of Politecnico di Milano and v)Rosa, by Medtech. The Neurobot and the Robocast roboticsystems are still in development stage.

We also found neurosurgical robotic systems built towardsother surgical procedures (like endoscopic handling, instru-mentation guidance and tele-operated actuation) that had thepotential to be used in a stereotactic surgery like DBS.Among those we selected: Evolution1, by Universal RobotSystems, Minerva from the Microengineering laboratory in theSwiss Federal Institute of Technology and NeuroArm from theUniversity of Carlgary. However, only the Evolution1 system iscurrently available at market and according to the informationwe have, the Minerva project has been discontinued.

Upon analyzing each oriented and adapted system featuresand based on the information gathered about DBS surgery,it was possible to outline a set of desired and unwantedcharacteristics for our project. However, all state of the artsystems presented one or another uninviting feature like:

1) Parallel manipulator arquitecture;2) Rely on stereotactic frame;3) Fixed robotic base, compromised mobility;4) Need of an integrated imaging machine (MRI/CT);5) Have integrated imaging/planning softwares;6) Complex system architectures;7) Exaggerated costs.We seek a serial actuator with 6 or more degrees of freedom

(DOF), for increased flexibility and broader workspace withthe ability to reach specific positions with no orientation re-striction. The robot should have stiff and lethargic movementsto avoid further stress/overheat in its joints and also to facilitate

motion stop action since the movement inertia is smaller.However, the choice of the robot is primarily conditionedby precision and repeatability since the slightest deviationfrom a target deep brain structure can reproduce very differentstimulation effects (reference).

In term of robotic system features, we concluded that itshould optimally include: i) a mobile platform, to be easilymoved inside or outside the operating room; ii) vision system,so the robot can recognize its position relative to the surgicalreferential and also to have feedback on the surroundingenvironment (avoid collisions); iii) have an integrated or’an interface with’ the imaging/planning software – whereneurosurgeons extrapolate surgical coordinates – used by thehealthcare institution; iv) Have simple, intuitive usage andkeep low acquisition and maintenance costs.

IV. INDUSTRIAL ROBOT SELECTION

Having defined a the sought robotic system characteristics,we searched the most renowned industrial robotic producers,for robotic systems that fit in the outlined profile. We selected25 serial manipulator systems from companies like: ABB,Adept, Epson, Fanuc, Kuka, Mitsubishi, Motoman, Nachi,Schunk, Staubli, Toshiba and Universal Robots.

Starting from 25 systems, we conducted several comparativeanalyses with the parameters displayed at the producers datasheets, such as controller and robot weight, horizontal reach,robot repeatability and payload capacity. We assigned a limitthreshold to each variable and in the end we reduced ourpotential robotic system pool from 25 to 3 robots from:

• ABB, a 6 DOF serial manipulator;• Motoman, a 6 DOF serial manipulator;• Schunk, a 7 DOF serial manipulator (also to assess the

impact of the extra degree of freedom).Other specifications like the manipulator dimensions,

structure and joint limits could not be directly comparedbetween systems, since its implication depends on the robotworkspace and in the task characteristics. To test each of thisvariables we implemented a virtual instance of each robot ina simulator (cf. section V).

Robotic tasks can be divided into small sequences of point-to-point or velocity/acceleration specified motions. The robotmanipulator movement is executed through the control of eachjoint along its kinematic chain. Therefore it is essential to mapthe relationship between the space coordinates where the robotoperates (Cartesian space) to the robot joints’ positions (Jointspace).

The problem of Direct Kinematics, expresses the forwardrelation between the robot joint values and the resultingcartesian coordinates. On the other hand, the problem ofInverse Kinematics expresses the set of joint values to achievea desired/input position, velocity or force in the cartesianspace. For the 3 selected manipulators, we developed theGeometric and Differential Kinematics for the Direct andInverse problems. Geometric Kinematics allowed us to controlthe robot based on cartesian and joint positions and were one

essential tool for positioning and holding surgery tasks. Dif-ferential Kinematics consider velocity and acceleration/forcevalues instead, and were used in the manipulation of surgi-cal instrumentation based in velocities and relative positions(straight line motion).

Detailed information about the Kinematics developed canbe found at [4].

V. IMPLEMENTED SOLUTION

At this point, we had developed a low level control al-gorithm based in desired positions and velocities for the 3candidate robotic systems. Further assessment of each manip-ulator should be carried either using the real robot or througha simulation environment. Since robotic manipulators are nota cheap and of easy access tool, we resorted to a roboticsimulator to test the kinematic equations and to implementthe control strategies oriented to the surgery tasks.

A. Robotics Simulator

The simulator should emulate the selected robots’ featuresand dimensions, and also the operating room environment. Tounderstand how to better fit the robot in its workspace and howto adapt the control approach to the surrounding elements, itis essential to recreate the operating room. Thus, we soughta open source software to implement each of the specificelements, either robots and virtual world. In addition to thecustom components, the simulator should include both graph-ical and physical representations of each virtual component.

According to the aimed features, considering the availablealternatives and the local shared knowledge, we chose to de-velop our solution in a 3D robotics simulator created in-house,the CoopDynSim. Besides gathering all the cited features, italso facilitates the code portability between simulation andreal robot, due to the abstraction of the communication layerbetween the control application and the robot either virtual orreal [10].

Fig. 2. CoopDynSim, operating room emulated environment.

CoopDynSim is built on C++, uses OpenGL graphical li-brary and runs on NewtonGD physics engine. It was originally

developed for mobile robots and basic shaped environments.Thus, one of our tasks involved importing and designing 3Dmodels of medical equipment from the Neurosurgery Serviceof Coimbra University Hospitals operating room (fig. 2).

The 3 selected industrial robots were also included in thesimulator. The 3D models of each robot were provided at theproducers website. We did slight changes to the 3D models inorder to remove superfluous details and parts and thus reducethe computational cost of drawing these models. We added aphysical representation to each link of each robot, who usedby the physics engine to emulate the mass, form, and physicalinteractions of each part.

At this point the robots were just a set of scattered links,which we had to connect using hinge physical joints accordingto the robot architecture. It was implemented virtual actuatorswith PID controllers that respond to desired joint position andvelocity inputs.

Fig. 3. ABB robot following a DBS surgery pre-defined trajectory.

We implemented specific DBS surgery end-effectors, in-cluding the trepan tool and electrode holding device, bothwith a linear actuator, to be moved independent from themanipulator. Furthermore, we created a feature called SurgicalPlan, which allows the simulator user to visualize the surgerytargets and trajectories in the virtual world (represented byphysicless, semi-transparent marks). This functionality facili-tated visualization of target and trajectories entities and aidedthe debug process of the control application, fig. 3.

B. Control Application

With the virtual operating room and robots, we starteddeveloping the control application. Given the abstraction pro-vided by the communication layer, we chose to implement thecontrol application in MatLab due to the quick, ease algorithmtesting and also to fulfill a request of the medical team we arecooperating with, since it is a programming language knownfrom part to part.

The control application was built in several modules/classes:i) Communication, ii) Robots, iii) DBS, iv) Kinematics and v)Utilities, all centered/connected to User-Interface (UI), fig. 4.

The control application establishes the communication be-tween the server ports associated to the robot modules, and theclient ports linked to the control application. In the Developer

Fig. 4. Control Application User-Interface.

panel, the application lets the user control individual robotjoints, control the robot by inputing a desired position andorientation and read, in real time, the robot end-effectorposition relative to a surgery defined referential.

The Clinical User panel has a feature that allows the user tomanage the surgery coordinates to be followed by the robot.Most of the actions in DBS surgery are performed along astraight line, from the skull entry point towards the target tobe stimulated. Thus, the user is asked to introduce these twosets of 3D coordinates for each target to be stimulated. Thecontrol application communicates with the simulator world,which creates the trajectory and the target marks relative tothe surgery referential. The clinical user can insert, removeone or several trajectories/targets.

When the user selects one trajectory/target to be stimulated,if there are other marks added to the Surgical Plan, they willbe hidden so the user knows exactly where the robot willactuate. From here we split the control approach algorithm intwo parts:

1) Positioning – where the robot will make the initialapproach to the linear trajectory of electrode insertion;

2) Manipulation – the robot will be locked to this lineartrajectory and will only execute upward or downwardmovements.

After selecting the desired trajectory, in the Positioningstage, the user should input the distance from the entry pointand along the trajectory, where the robot end-effector shouldposition. We also implemented a function to compute andshow in real time the distance from the robot end-effectorto the stimulation target.

At the Manipulation stage, the robot can only move alongthe locked trajectory. The control application starts by identi-fying what end-effector is attached to the hand of the manip-ulator (if any). The robot is expected to execute incremental

movements toward or away from the target, that can eitherbe accomplished by a joint action of the whole manipulator– resorting to Differential Kinematics – or by executingindependent increments of the instrumentation (trepan andelectrode drive) relative to the robot arm.

Safety is currently pointed as the most discouraging fac-tor for the use of robotic instrumentation within operatingrooms [8]. As such an important factor, we established safetyroutines in the control application. Firstly, motion restrictionwas created so the robot does not deviate from an assignedlinear trajectory, thus preventing inadvertent movements. Ad-ditionally, when the trepan tool is attached to the manipulatorhand, the robot stops any motion that would cause the drillto pass behind the skull entry point, to avoid brain damage.Precision control routines, compute the distance between thedesired and the generated solution, position and orientation.If the generated position and orientation fall outside a safetythreshold distance, the robot instead of moving, stops andnotifies the user3. We also created a log file system, so thecontrol application registers every user actions, variables andintroduced values. Each log file is associated to a uniquesurgical procedure and it is registered the time (hh:mm:ss)when each event happened.

VI. RESULTS

At this stage, we want to evaluate the performance of thedeveloped system mainly in two areas: overall precision andtask suitability/system flexibility. It was impossible to assessthe precision of the selected robots and control approach, dueto the unavailability of information regarding the robot jointcontrollers and manipulator precision.

However, we assessed each robot flexibility and suitabilityperformance towards the required task, taking into accounttheir structure and possible conflicts with the surroundingoperating room environment. To do so, we developed a simpletask where we tested several combinations of: robot type;robot base platform heights, relative to the patient’s head;robot orientation, relative to the patient’s sagittal plane; for14 generic trajectories depict in, fig. 5.

Fig. 5. Selected robots flexibility test.

3These safety checks are performed before any action (Positioning andManipulation stages).

Each cell of 6a, 6b or 6c represents a set of 14 trajectoriesfor a specific robot height (of robot’s end-effector relative tothe patient’s head height, in home position) and orientation (ofrobot’s home position posture relative to the patient’s sagittalplane – 0o when the robot is in front of the patient).

(a) ABB robot flexibility results.

(b) Motoman robot flexibility results.

(c) Schunk robot flexibility results.

Fig. 6. Flexibility test results for several robot base heights and robotorientation relative to the patient. Green, robot can reach 150mm from theskull entry point along the insertion trajectory and move towards the target.Yellow, robot can reach Xmm from the skull entry point along the insertiontrajectory and move towards the target. Orange, robot can reach the insertiontrajectory but collides with end-effector instrumentation. Red, robot can notreach the insertion trajectory.

Comparing the results of both 6 DOF manipulators (ABBrobot and Motoman robot) we see that for starting end-effector

heights equal or below the patient’s head the robot fails toreach several trajectories, even the most probable ones (6central trajectories). For the remaining heights (+100mm,+200mm and +300mm), both robots can reach most ofthe defined trajectories, and with success the most probableones. However, we suggest as preferable starting end-effector’sheight +200mm, because for +100mm the trajectories on theopposite side of the robot manipulator relative to the patient’ssagittal plane are reached at the limit of the robot’s workspace,which can comprise its stability and precision. At +300mmheight, the robot collides with the end-effector instrumentationfor the most central trajectories for greater orientation angles.

We point as preferable robot orientation angle between+15o and +30o, since the robot achieves better results forsmaller angles. At the same time, the robot occupies more ofthe neurosurgeon’s workspace for smaller orientation angles.

On the other hand, the 7 DOF Schunk robot presents thebest flexility results, followed by Motoman robot with ABBrobot, which shows the worst flexibility among the selectedsystems. The extra DOF and the larger arm horizontal reach,enable the Schunk robot to successfully reach most of thetrajectories for almost every combination of robot base heightand orientation. Thus, if workspace conditions are a problem,a 7 DOF robotic arm is a option to consider.

However, the 7 DOF manipulator, unlike the others, collidedseveral times with the surrounding equipment, mainly due tothe elbow redundancy caused by the extra DOF4. Furthermore,the extra joint in a serial manipulator will add another errorsource to the end-effector final position and orientation. Hencethe need to further invest in joint actuators precision, whichwill ultimately enhance the final product cost.

For these reasons and facing the results achieved so far, wedare to point Motoman robot as the most fit robotic systemfor DBS surgery assistance.

VII. CONCLUSIONS

The knowledge acquired about DBS surgery, the closecontact with field professionals – which may later turn to end-users of this project – and by witnessing what was alreadyachieved with robotic systems in stereotactic neurosurgery,allowed us to outline a set of desirable features for the soughtrobotic system. We compiled information about the existingmass produced industrial robots, to select the most suited forour purpose. Based on the selected systems, we developedspecific control algorithms oriented to DBS surgery tasks,and created several tools to test and consolidate the controlapplication. With these tools, it was possible to assess eachmanipulators’ flexibility, suitability to the assigned tasks, andto analyze their interaction with the operating room medicalequipment.

In terms of future work, it includes devising a mobileplatform so the robot can be easily moved inside and outsidethe operating room and an attach system to fix the mobile

4The control application cost functions, still don’t include collision avoid-ance routines. Nonetheless, it was evident the disparity between the numberof collisions occurring between 7 and 6 DOF manipulators.

platform to the patient’s reference stereotactic system. Anothergoal to achieve is a system to recognize the robot’s positionand orientation relative to the surgical referential, as thistransformation isn’t static.

Most importantly, the information gathered so far reinforcedour perspectives about the viability of this new and promisingproject, due to the availability of the resources needed toachieve a final solution and to the fact that it gathers a setof attractive features, not yet explored in the current marketoffer.

ACKNOWLEDGEMENTS

This work has been partially financed by projectsFP7 Marie Curie ITN - NETT (project no289146), FCTFCOMP-01-0124-FEDER-022674, Pest-C/MAT-UI0013/2011(FCT grant ref. UMINHO/BIC/8/2012) and FCT PhD grant(ref. SFRH/BD/86499/2012).

REFERENCES

[1] A. L. Benabid. Deep brain stimulation for Parkinson’s disease. CurrentOpinion in Neurobiology, 13(6):696–706, 2003.

[2] D. B. Camarillo, T. M. Krummel, and J. K. Salisbury. Robotictechnology in surgery: past, present, and future. American journal ofsurgery, 188(4A Suppl):2S–15S, October 2004.

[3] F. Cardinale and R. Mai. Robotic Implantation of Intracerebral Elec-trodes in Epilepsy Surgery. Congress of Neurosurgical Surgeons Spring,12(2):24–26, 2011.

[4] C. Faria. Robotic Implantation of Intracerebral Electrodes for DeepBrain Stimulation. Master thesis, University of Minho, 2012.

[5] T. Haidegger, L. Kovacs, G. Fordos, Z. Benyo, P. Kazanzides, A. Kata-shev, Y. Dekhtyar, J. Spigulis, and R. Magjarevic. Future Trendsin Robotic Neurosurgery. In Alexei Katashev, Yuri Dekhtyar, andJanis Spigulis, editors, 14th Nordic-Baltic Conference on BiomedicalEngineering and Medical Physics, volume 20 of IFMBE Proceedings,pages 229–233, Riga, 2008. Springer Berlin Heidelberg.

[6] C. S. Karas and E. A. Chiocca. Neurosurgical robotics: a review ofbrain and spine applications. Journal of Robotic Surgery, 1(1):39–43,2007.

[7] L. Lau and M. Breteler. Epidemiology: Parkinson disease. The LancetNeurology, 5:525–35, 2006.

[8] S. Lavallee, J. Troccaz, L. Gaborit, P. Cinquin, A. L. Benabid, andD. Hoffmann. Image guided operating robot: a clinical applicationin stereotactic neurosurgery. In Proceedings 1992 IEEE InternationalConference on Robotics and Automation, pages 618–624, Nice, 1992.IEEE Comput. Soc. Press.

[9] E. D. Louis and R. Ottman. How common is the most common adultmovement disorder? Estimates of the prevalence of essential tremorthroughout the world. Movement disorders, 13(1):5–10, 1998.

[10] T. Machado, M. Sousa, S. Monteiro, and E. Bicho. CoopDynSim: a 3Drobotics simulator. In Robotica, pages 45–50, Braga, 2012. Universityof Minho.

[11] Erwin B Montgomery Jr. and John T Gale. Mechanisms of action ofdeep brain stimulation (DBS). Neuroscience Biobehavioral Reviews,32(3):388–407, 2008.

[12] N. Nathoo, M. C. Cavuolu, M. A. Vogelbaum, and G. H. Barnett.In touch with robotics: neurosurgery for the future. Neurosurgery,56(3):421–33; discussion 421–33, March 2005.

[13] Aidan Neligan and J. W. Sander. The incidence and prevalenceof epilepsy. In Epilepsy 2011, volume 76, chapter 1, pages 1–7.International League Against Epilepsy’s 13th, London, 2011.

[14] J. S. Perlmutter and J. W. Mink. Deep brain stimulation. Annual reviewof neuroscience, 29(1):229–57, January 2006.